I can not decipher recommendations for setting time for those with ppp
connections that can get very good time with ntpdate (referencing 3 ntp
servers). In the following, "hardware-clock" refers to the medium grade
clock that runs whether the computer is on or off; "Linux-clock" refers to
the time Linux has, which is very good because the CPU keeps very good time,
but the Linux clock dissappears on rebooting.
Presume the ppp connection will not go on at boot, but is automated for
connection sometime during the night. My impression is that the
recommendation would be to use an ip-up entry like
/usr/sbin/ntpdate -t 8 -b -s black-ice.cc.vt.edu ntp.css.gov louie.udel.edu
followed later by an ip-down entry like
/sbin/hwclock --set --date="`/bin/date`" --utc
which sets both the hardware-clock and the drift /etc/adjtime. But when we
reboot we want to immediately set the correct time, so in the absense of a
start-up ppp connection, we want the proper drift, set before the last
shutdown through "hwclock --adjust --utc". This is what I will try.
The package adjtime appears to try keeping correct time without changing the
hardware-clock. But again, with a ppp connection, one can presume we want the
correct time set at boot time through the hardware-clock and drift information
from /etc/adjtime. Then sometime while up, we might connect by ppp, getting
the correct time with ntpdate. So, adjtime should not change the current
Linux-time, but only maintain the drift, I gather through /etc/adjtime.
However, this leaves adjtime subject to something mucking with the
hardware-time; eg, "date" or Microsoft.
Here is how I feel time ought to work. Linux should not muck with the
hardware-clock, but only refer to it. Then, hopefully, any Microsoft
adjustments for dual-boot computers will be stable. This would require
absolute Linux-clock adjustments and drift adjustments to accomodate the
medium-grade hardware-clock. So, Linux could occasionally set the absolute
adjustment with ntpdate and later make drift adjustments for the
hardware-clock using the Linux-system clock (no need for another ntpdate).
This all presumes that while the system is up that the Linux-system clock need
not be adjusted for drift; the Linux system need only prepare drift estimates
for reboot and the medium-grade hardware-clock. All this amounts to is a 9th
grade algebra equation,
actual-time = hardware-time + absolute-adjustment-for-hardware-time +
time-since-last-absolute-adjustment * drift .
The hardware-time never changes, but we occasionally set
absolute-adjustment-for-hardware-time and estimate the drift.
Is there a way to do this, perhaps with adjtime? Or is there another
recommended way for ppp connections that gets the correct time with ntpdate?
--
Jim Burt, NJ9L, Fairfax, Virginia, USA
[EMAIL PROTECTED] http://www.mnsinc.com/jameson
[EMAIL PROTECTED]
"If merely 'feeling good' could decide, drunkenness would be the supremely
valid human experience." --William James, Varieties of Religious
Experience
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
[EMAIL PROTECTED] .
Trouble? e-mail to [EMAIL PROTECTED] .
